Building on the rails

I started with a romantic idea: a parallel stack. Something built alongside the country's digital public infrastructure — its own protocols, its own rails, a quiet second system that did things more humanely than the first. The research killed that idea, gently and completely. It deserved to die.

Every component of India's public stack that earned legitimacy did so from inside the existing order — as a statutory body, a regulated not-for-profit, a philanthropically-stewarded protocol, or a self-regulatory organisation that co-evolved with its regulator. There is no surviving example of a first-time builder standing up a genuinely parallel protocol and having it last. "Parallel," as a technical claim, has no precedent that held.

So the reframe — the one that actually lands — is this: not a parallel stack, but an application and governance layer riding the rails that already exist. Identity, credentials, discovery, consent: those are public infrastructure, and you build on them, not beside them. "Parallel" survives only as a governance claim — a separately-held stewardship of ethics and community value — never as a technical one.

The number that did the reframing

One public statistic reorganised the whole picture. India's flagship school-volunteering platform reports something like 8.5 lakh schools and 5 lakh registered volunteers — and, across more than four years, roughly fifty-six thousand completed service activities. That's about 0.015 activities per registered volunteer per year. It is registration theatre, not engagement.

That single number says two things at once. The opportunity is enormous. And a platform, by itself, will not touch it. Whatever gets built cannot be another place to register — it has to be the recognition, matching, and relationship layer that makes participation actually happen, and actually count. Federated into what exists. Not a replacement for it.

Credential sovereignty, or: write to the ledger

In India, the right to recognise learning is held by the state. There is a live, population-scale credit ledger now — academic IDs, a national credit framework, verifiable credentials at scale. The blunt consequence: anything that does not write to that ledger is a credential graveyard. A learner's effort that cannot be redeemed anywhere real is, at the point of redemption, worthless to them — however lovely the experience felt. Federation is not a nice-to-have. It is the difference between a keepsake and a credential.

Reading policy as a map

The most useful discipline of the whole exercise was treating education policy not as a wall but as a map. The national policy is large; the part of it that is actually operationally live is much narrower. So you read it for the clauses that already authorise the thing you want to build — vocational and open learning, lifelong learning, experiential pedagogy, holistic assessment — and you rank candidate ideas by how squarely existing policy already blesses them. The strongest idea is the one where the policy text is already waiting for someone to show up and do the work. Build there first. Everything else arranges itself in rings around that.

Precedents worth stealing from

The closest working analogues aren't the famous knowledge commons. They're quieter:

• A recognition economy — the donor-classroom model, where a specific person makes a specific ask, someone answers, and there's a tangible receipt and a thank-you loop. Except here the currency is time and skill, not money.
• Commons governance with teeth — Elinor Ostrom's eight design principles as a literal test: defined boundaries, locally-fitted rules, collective choice, monitoring, graduated sanctions, cheap conflict resolution, recognition by higher authority, nested governance. If a platform doesn't map to those eight, it isn't a commons, whatever its rhetoric says.
• The alumni multiplier — fellowship models where graduates keep reaching learners for decades, and a meaningful fraction go on to found things themselves.
• Narrative as mobiliser — the single-photographer, single-story model that moved real money and attention with nothing but attention paid well.

And one cautionary thread worth its own paragraph. The dream of "we'll also be a media house" is, at seed stage, almost always a distraction. Lavishly-funded editorial ventures have folded inside two years; the publications that endure are subsidised indefinitely by a profitable core that has no user-acquisition funnel they're trying to replace. The honest version at the start is not a media house — it's the founder as the medium: write in public, extract the stories from the work itself, let the compounding do the rest.

The contribution I'm proudest of

Site Reliability Engineering — the discipline that keeps large systems alive — has a quietly beautiful operational culture: blameless postmortems, observability, relentless reduction of toil, a rhythm of flow, feedback, and learning. Monitoring, evaluation and learning in the social sector is suffocating under funder-report toil and would be transformed by that culture. The bridge between the two is real, and as far as I can tell, no one had written it down as an integrated claim.

But it breaks in one specific, important place. SRE has the concept of an error budget — a permitted, expected amount of failure. You cannot have an error budget for human welfare. You cannot "budget" for a child being harmed. So the genuinely novel piece is the replacement: an Ethical Error-Budget Wall, a taxonomy that separates three kinds of metric and treats them completely differently —

• Zero-tolerance — safeguarding, child harm. No budget at all. Any incident triggers an unconditional stop-and-review.
• Operational — uptime, onboarding, data quality. Classic error budgets apply; failure is expected and bounded.
• Outcome — learning gains, whether a match survived. "Learning budgets," with reserved room for honest, pre-registered experiments.

The engineering rigour ports. The thing it must never be allowed to port — budgeted harm to vulnerable people — gets walled off explicitly, by name, at the top of the system. That wall is the point.

Ethics as a product decision, not a press release

The temptation is to write a grand ethics charter — a "Belmont for industry." The evidence is brutal on this: voluntary ethics frameworks from non-government actors, absent regulation or certification or market pull, reliably fail to bind behaviour. They drift the moment they're inconvenient. So the discipline is the opposite of grandeur. Ethics shows up as product decisions you can be audited on — no behavioural targeting of minors; an explicit, visible way of attributing community contributions back to the people who made them; logs that are ready to be inspected. Make the commitments falsifiable. Publish what you actually do, not what the industry should.

How I tried to hold it

I wrote the whole brief to be wrong on purpose — every major claim paired with the evidence that would falsify it. That habit is the real deliverable. The romantic version of this venture wanted to be parallel. The honest version is an application layer on open rails, with its parallelism reserved for the one place it's defensible: the stewardship of ethics and community value that no one else is holding.

Parallel yields. Blessing compounds. Architectural purity does not feed a single student — federation does.

That sentence cost me the most romantic part of the idea. It kept everything I actually wanted to build.